When testing for WEB system performance, the first time a test is thought WEB system can withstand a maximum concurrent virtual user (VU) number of users, because the maximum number of concurrent systems VU may be reacted directly carrying capacity of the system. But people often ignore the concurrency model of VU. What is the VU concurrency model it? As shown below, we use a browser to access a page, the browser will have a plurality of HTTP requests sent to the server, these requests also have a serial parallel (water requests overlap in time concurrent requests belonging serial request refers to: before the end of a request, the request only after a request is issued). If the performance test tool provides WEB recording capabilities and in accordance with the behavior model to simulate the behavior of the browser VU, it is the best (if you use Jmeter or LoadRunner 11 then the tool can not be done, see: " Performance testing tools Jmeter you do not know the inside story , "" performance testing tool LoadRunner inside you do not know "). Up to now I understand the performance testing tool: kylinTOP can be achieved (Jmeter, CPTS (Huawei), PTS (Ali) were unable to achieve), but the article does not discuss how to use kylinTOP to record a script and simulate browser behavior, kylinTOP is how to use hand-build such a scenario. Why build it by hand, mainly some web services provided by the external system is not a WEB page, but HTTP interface functions, for this scenario we need to build by hand.
If the tested system is a WEB service system, provide both HTTP interface is not a WEB page, but HTTP interface functions. For such a system performance test how we should do? To test the maximum number of concurrent VU can bear, we must first analyze its model of VU's request. VU request model assumed as follows:
VU request Model Description: WEB system provides a known A, B, C, D, E, F, G, H of the upper layer user, I request a total of nine interfaces. An upper layer user requests nine constraint requirement is A, B is issued at a serial, C, D, E, F are four request issued in parallel, G, H, I are made under serial. Falls under the transmission request according to the following pattern:
A top user of WEB systems is a VU, then how to manually build such a model request it?
Step 1: Create a new script
Step 2: Create the node requesting node and description
By "add" button to add HTTP request node and the node description. While concurrent requests to the mobile node to be described below (described as a child node, while the node name description l Type: start_concurrent)
Description: HTTP request manually add kylinTOP festival node default is serial, "describes the node" a child node process is concurrent requests (node name must be described: start_concurrent, or else as serial processing).
Step 3: Modify the script property
The setup script "Advanced Settings - virtual user" sign under any graph attribute value selection.
Step 4: Debugging Scripts
Click the debug button to generate VU falls as shown in FIG. The falls figure: CDEF concurrent requests is four, the remaining serial request, consistent with the requirements of the scene.
Here, waterfall scene with a script that runs requirements have been exactly the same. The subsequent use of the new test script to test the task to run the maximum number of VU get the system can bear. If Jmeter, PTS and other performance testing tools to test, the VU HTTP request is a linear model request (request serial), can not generate the required scene description script, the test results are not very accurate.